Abnormality determination device, abnormality determination method, and storage medium

Abstract

An abnormality determination device, an abnormality determination method, and a storage medium for a water pump are provided. An input variable mapped by processing image data obtained by capturing an outer surface of the water pump is acquired. A circuit executes a plurality of provisional determination results from a plurality of mappings. Each of the plurality of provisional determination results is obtained by executing a provisional determination process for each of output variables output from the plurality of mappings. A determination determination process determines provisional determination results that account for more than half of the plurality of provisional determination results as a final determination result of whether a leakage of cooling water has occurred in the water pump.

Description

Technical Field

[0001] This invention relates to an anomaly detection device, an anomaly detection method, and a storage medium. Background Technology

[0002] Japanese Patent Application Publication No. 2004-108250 discloses an internal combustion engine equipped with a water pump. The water pump includes a pump casing, a pump shaft, an impeller, and sealing components. The pump casing divides a flow space for cooling water to circulate. The pump shaft passes through the pump casing and is rotatably supported relative to the pump casing. A portion of the pump shaft, including its first end, is located within the flow space. The impeller is fixed to the first end of the pump shaft. By rotating together with the pump shaft, the impeller pressurizes the cooling water within the flow space of the pump casing to various parts of the internal combustion engine.

[0003] The sealing component is mounted on the outer circumferential surface of the pump shaft. The sealing component is located on the second end side of the pump shaft, including the first end, when viewed from the impeller. The sealing component prevents cooling water from leaking from the flow space of the pump casing to the outside of the pump casing. Summary of the Invention

[0004] The problem that the invention aims to solve

[0005] In water pumps like the one described above, excessive cooling water may leak from the flow space of the pump casing to the outside, due to factors such as deterioration of sealing components. Traditionally, during pump maintenance, operators would observe the outer surface of the pump casing to determine if a cooling water leak had occurred. However, this method of assessment can be biased due to the subjective judgment of each operator.

[0006] Technical solutions for solving the problem

[0007] According to one aspect of the present invention, an anomaly determination device is provided. The anomaly determination device uses a water pump that pumps cooling water for an internal combustion engine as the determination object. Based on image data obtained by photographing the outer surface of the water pump, the anomaly determination device determines whether a leak of cooling water has occurred in the water pump. The anomaly determination device includes an execution circuit as an actuation device and a storage circuit as a storage device. The storage circuit stores mapping data with defined mappings. The mapping outputs an output variable indicating whether a leak of cooling water has occurred in the water pump by inputting an input variable. The execution circuit performs: an acquisition process, which acquires the input variable from the image data; and a calculation process, which outputs the value of the output variable by inputting the input variable acquired by the acquisition process into the mapping. The execution circuit performs the following processes: a temporary determination process, which performs a temporary determination on whether a leak of cooling water has occurred based on the output variable; and a determination confirmation process, which performs a final determination on whether a leak of cooling water has occurred based on the temporary determination result as a determination result of the temporary determination process. The mapping data defines multiple mappings that are different from each other. One or more of the multiple mappings are mappings that have been learned in advance through machine learning. The temporary determination process is one of a plurality of temporary determination processes. The temporary determination result is one of a plurality of temporary determination results. The execution circuit performs the calculation process for each of the plurality of mappings, and performs the temporary determination process for each of the output variables output from the plurality of mappings. In the determination process, the execution circuit takes the temporary determination results that account for more than half of the temporary determination results of the plurality of temporary determination processes as the final determination result of whether the cooling water leakage has occurred.

[0008] Based on the above structure, a determination of whether a cooling water leak has occurred is made according to the mapping specified by the mapping data, based on the input variables obtained from the image data. Since there is no room for subjective intervention by the operator in this series of determinations, the determination result will not be biased due to the subjectivity of each operator. Furthermore, according to the above structure, the final determination result is not based on a single mapping, but rather on more than half of the results from provisional determinations based on multiple mappings. Therefore, the accuracy of the determination result can also be ensured.

[0009] In the above structure, the mapping that uses the most types of variables as input variables among the multiple mappings is a specific mapping. Alternatively, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that a cooling water leak has not occurred, the execution circuit, in the determination process, uses the temporary determination result from the temporary determination process based on the output variable output from the specific mapping as the final determination result of whether a cooling water leak has occurred.

[0010] In the above structure, the more types of input variables that accompany the input to a specific mapping, the higher the reliability of the provisional judgment result based on the output variable output from the specific mapping. According to this structure, when the number of provisional judgment results indicating a cooling water leak is equal to the number of provisional judgment results indicating no cooling water leak, the more reliable provisional judgment result is adopted. Therefore, even when the number of conflicting provisional judgment results is high, the provisional judgment result considered more accurate can be used as the final judgment result.

[0011] In the above structure, multiple mappings may also be pre-learned through ensemble learning based on the same learning data.

[0012] Based on the above structure, for example, compared to the case of learning multiple mappings based on learning data obtained under different conditions, it is possible to suppress bias in the judgment results.

[0013] According to another aspect of the invention, a non-transitory, computer-readable storage medium is provided that stores an anomaly determination program for causing an execution circuit to perform anomaly determination processing. The anomaly determination processing takes a water pump that pumps cooling water for an internal combustion engine as the determination object. Based on image data obtained from photographing the outer surface of the water pump, the anomaly determination processing determines whether a leak of the cooling water has occurred in the water pump. The storage circuit has mapping data with defined mappings. The mapping takes input variables as inputs and outputs an output variable indicating whether a leak of the cooling water has occurred. The anomaly determination processing includes the following processes, performed by the execution circuit: an acquisition process, which acquires the input variables from the image data; and a calculation process, which outputs the value of the output variable by inputting the input variables acquired by the acquisition process into the mapping. The anomaly determination processing includes the following processes, performed by the execution circuit: a temporary determination process, which performs a temporary determination on whether a leak of the cooling water has occurred based on the output variable; and a determination finalization process, which performs a final determination on whether a leak of the cooling water has occurred based on the temporary determination result as a determination result of the temporary determination process. The mapping data defines multiple mutually different mappings. One or more of the plurality of mappings are mappings that have been pre-learned through machine learning. The temporary determination process is one of a plurality of temporary determination processes. The temporary determination result is one of a plurality of temporary determination results. The anomaly determination process includes the following steps: the execution circuit performs the computation process for each of the plurality of mappings; and the temporary determination process is performed for each of the output variables output from the plurality of mappings. The anomaly determination process includes the following step: the execution circuit, in the determination determination process, takes the temporary determination results from the plurality of temporary determination processes that account for more than half as the final determination result for whether the cooling water leakage has occurred.

[0014] Based on the above structure, the presence or absence of a cooling water leak is determined according to the input variables obtained from the image data and the mapping specified by the mapping data. There is no room for subjective intervention by the operator in this series of determinations. Therefore, the determination result will not be biased due to the subjectivity of each operator. Furthermore, according to the above structure, more than half of the provisional determination results based on multiple mappings, rather than a single mapping, are used as the final determination result. Therefore, the accuracy of the determination result can also be ensured.

[0015] In the above structure, the mapping with the most types of variables used as input variables among the multiple mappings is designated as a specific mapping. Alternatively, in the anomaly determination process, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that a cooling water leak has not occurred, the execution circuit, in the determination process, uses the temporary determination result from the temporary determination process based on the output variable output from the specific mapping as the final determination result for whether a cooling water leak has occurred.

[0016] In the above structure, the more types of input variables that accompany the input to a specific mapping, the higher the reliability of the provisional judgment result based on the output variable output from the specific mapping. According to this structure, when the number of provisional judgment results indicating a cooling water leak is equal to the number of provisional judgment results indicating no cooling water leak, the more reliable provisional judgment result is adopted. Therefore, even when the number of conflicting provisional judgment results is high, the provisional judgment result considered more accurate can be used as the final judgment result.

[0017] According to other aspects of the invention, it can also be embodied as an anomaly determination method that performs various processes described in relation to any of the above-described anomaly determination devices.

[0018] According to other aspects of the invention, it can also be embodied as a non-transitory computer-readable recording medium storing a program that causes a processing device to perform various processes recorded in relation to any of the aforementioned anomaly detection devices. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the internal combustion engine.

[0020] Figure 2 It is aimed at Figure 1 A schematic diagram of the abnormality detection device for an internal combustion engine.

[0021] Figure 3 It means used for Figure 2 The flowchart shows the steps of the anomaly detection device in making a determination.

[0022] Figure 4 It means Figure 3 The flowchart for the first temporary decision control in the process.

[0023] Figure 5 It means Figure 3 The flowchart for the second temporary decision control.

[0024] Figure 6It means Figure 3 The flowchart for the third temporary decision control.

[0025] Figure 7 It means Figure 3 The flowchart for the fourth temporary decision control. Detailed Implementation

[0026] It should be understood that "at least one of A and B" as described in this specification means "A only", "B only" or "both A and B".

[0027] <Brief Structure of an Internal Combustion Engine>

[0028] The following is based on Figures 1 to 7 One embodiment will be described. First, the general structure of the internal combustion engine 100 of the vehicle will be described. Furthermore, in the following description, when the vertical direction is mentioned, it refers to the direction observed from the driver sitting in the driver's seat of the vehicle with the internal combustion engine 100 mounted on it.

[0029] like Figure 1 As shown, the internal combustion engine 100 includes a cylinder block 10, a water pump 20, a bracket 30, and a pulley 40. Although not shown in the diagram, the cylinder block 10 is divided into multiple cylinders. Furthermore, the cylinder block 10 has an internal space 11 separate from each cylinder. The internal space 11 is the flow path for cooling water used to cool the internal combustion engine 100. A portion of the internal space 11 opens into the side wall of the cylinder block 10.

[0030] The water pump 20 includes a pump housing 21, a pump shaft 22, an impeller 23, a bearing 24, a sealing component 25, and a plug 26. The pump housing 21 is fixed to the side wall of the cylinder body 10. The pump housing 21 covers the opening of the internal space 11 of the cylinder body 10. Therefore, the pump housing 21 and the cylinder body 10 together define a flow space 100Z for the flow of cooling water. In this embodiment, the cooling water is pink.

[0031] The pump housing 21 has a through hole 21A. A bearing 24 is located within the through hole 21A. The bearing 24 rotatably supports the pump shaft 22 relative to the pump housing 21. The pump shaft 22 is generally rod-shaped. The first end of the pump shaft 22 is located at... Figure 1 The middle is the left end, and the second end of the pump shaft 22 is at... Figure 1The middle section is the right end. A portion of the pump shaft 22, including its first end, is located within the flow space 100Z. The impeller 23 is fixed to the first end of the pump shaft 22. By rotating together with the pump shaft 22, the impeller 23 pressurizes the cooling water within the flow space 100Z to various locations. A sealing member 25 is installed on the outer circumferential surface of the pump shaft 22. The sealing member 25 is located on the second end side of the pump shaft 22, including its first end, when viewed from the impeller 23. Furthermore, the sealing member 25 is located at the portion of the through hole 21A closest to the flow space 100Z. The sealing member 25 prevents cooling water from leaking from the flow space 100Z into the through hole 21A. That is, the sealing member 25 prevents cooling water within the flow space 100Z from leaking to the outside of the pump casing 21.

[0032] A portion of the pump shaft 22, including its second end, protrudes outward from the pump housing 21. A pulley 40 is fixed to the second end of the pump shaft 22 via a bracket 30. The pulley 40 is connected to the crankshaft of the internal combustion engine 100 via a belt (not shown). Therefore, the pulley 40 rotates under the driving force from the crankshaft of the internal combustion engine 100. Furthermore, the rotation of the pulley 40 causes the pump shaft 22 and the impeller 23 to rotate.

[0033] Pump housing 21 includes an upper space 21B, an upper passage 21C, a lower space 21D, and a lower passage 21E. The upper space 21B is located on the upper side of the pump housing 21 when viewed from the through hole 21A. A portion of the upper space 21B opens into the outer wall of the pump housing 21. The upper space 21B is connected to the through hole 21A via the upper passage 21C. The lower space 21D is located on the lower side of the pump housing 21 when viewed from the through hole 21A. A portion of the lower space 21D opens into the outer wall of the pump housing 21. The lower space 21D is connected to the through hole 21A via the lower passage 21E. A plug 26 blocks the opening of the lower space 21D. The plug 26 restricts the leakage of cooling water to the outside of the pump housing 21 via the lower space 21D.

[0034] In the water pump 20, even if the sealing component 25 is functioning normally, cooling water that has changed into gas may sometimes leak from the flow space 100Z to the through hole 21A through the sealing component 25. Thus, the gaseous cooling water reaching the through hole 21A may leak to the outside of the pump casing 21 via the upper passage 21C and the upper space 21B. Furthermore, when the gaseous cooling water reaching the through hole 21A is cooled, it changes into liquid. This liquid cooling water may leak to the outside of the pump casing 21 via the lower passage 21E and the lower space 21D. However, since the amount of such cooling water leakage is very small, the aforementioned leakage is permissible in the design.

[0035] <Simplified Structure of the Anomaly Detection Device>

[0036] Next, the anomaly determination device 200, which uses the water pump 20 as the determination object, will be described. In this embodiment, the anomaly determination device 200 is used in places where vehicle maintenance is performed, such as automobile repair shops.

[0037] like Figure 2 As shown, the anomaly detection device 200 includes a camera 210, a display 220, and a control device 290. The camera 210 is capable of capturing images of a subject. The display 220 is capable of displaying various information. In this embodiment, the display 220 is a so-called touch panel display. That is, the display 220 can also accept input of various information.

[0038] The control device 290 is electrically connected to the camera 210 and the display 220. Therefore, the control device 290 is able to acquire image data captured by the camera 210. In addition, the control device 290 can enable the display 220 to display various information.

[0039] The control device 290 includes a CPU 291, peripheral circuitry 292, ROM 293, storage device 294, and bus 295. Bus 295 communicatively connects the CPU 291, peripheral circuitry 292, ROM 293, and storage device 294. ROM 293 pre-stores various programs for enabling the CPU 291 to function as a processing device performing various controls. Storage device 294 pre-stores mapping data 294A. Mapping data 294A defines multiple distinct mappings. The mappings defined by mapping data 294A are input variables, thereby outputting an output variable indicating whether a cooling water leak has occurred in the water pump 20. In this embodiment, mapping data 294A defines a first mapping M1, a second mapping M2, a third mapping M3, and a fourth mapping M4. The first mapping M1 is a so-called relation. The second mappings M2 to the fourth mapping M4 are mappings pre-learned through machine learning. Further details regarding the first mapping M1 to the fourth mapping M4 will be described later. The peripheral circuitry 292 includes circuits for generating clock signals for predetermined internal operations, power supply circuits, reset circuits, etc. In this embodiment, the CPU 291 and ROM 293 constitute an execution device or execution circuit. Additionally, the storage device 294 constitutes a storage device or storage circuit. Furthermore, an example of the anomaly detection device 200 is a smartphone, which functions as a computer. By executing the anomaly detection program, anomaly detection processing, or anomaly detection method pre-stored in the ROM 293 and storage device 294 through the execution circuit, the smartphone functions as the anomaly detection device 200.

[0040] <Steps for determination>

[0041] Next, the steps by which the anomaly detection device 200 determines whether a coolant leak has occurred in the water pump 20 will be explained. Furthermore, this detection step is performed, for example, during maintenance of the water pump 20 in an auto repair shop.

[0042] like Figure 3 As shown, in step S11, a user, such as an operator, uses the camera 210 of the anomaly detection device 200 to take pictures of the area around the opening of the lower space 21D in the outer wall surface of the water pump 20, i.e., near the plug 26. Additionally, the operator notifies the anomaly detection device 200 that the recording of the water pump 200 has ended by operating an icon on the display 220 of the anomaly detection device 200. Then, the control device 290 of the anomaly detection device 200 acquires the image data captured by the camera 210. Afterwards, the control device 290 causes the processing to proceed to step S12.

[0043] In step S12, the control device 290 performs an acquisition process to obtain input variables from the image data captured in step S11. Then, the control device 290 performs a calculation process to output the value of an output variable by mapping the input variables obtained through the acquisition process into the mapping data 294A. Furthermore, the control device 290 performs a temporary determination process to temporarily determine whether a cooling water leak has occurred based on the output variables. Here, the control device 290 performs a calculation process for each of the multiple mappings of the mapping data 294A, and performs a temporary determination process for each output variable output from the multiple mappings. As described above, the mapping data 294A specifies a total of four mappings, from the first mapping M1 to the fourth mapping M4. Therefore, in this embodiment, the control device 290 performs a total of four temporary determination processes using the first mapping M1 to the fourth mapping M4. The specific details of the processing in step S12 will be described later. Afterward, the control device 290 proceeds to step S13.

[0044] In step S13, the control device 290 performs a final determination process to determine whether a cooling water leak has occurred, based on the preliminary determination results of the preliminary determination processes. In this determination process, the control device 290 uses the preliminary determination results from multiple preliminary determination processes that constitute more than half of the preliminary determination results as the final determination result for whether a cooling water leak has occurred. For example, if a cooling water leak is determined to have occurred in three out of four preliminary determination results, the control device 290 ultimately determines that a cooling water leak has occurred. Conversely, if a cooling water leak is determined not to have occurred in three out of four preliminary determination results, the control device 290 ultimately determines that a cooling water leak has not occurred. Furthermore, for example, if the number of preliminary determination results indicating a cooling water leak is the same as the number indicating no cooling water leak, the control device 290 uses the preliminary determination result based on the output variable output from the fourth mapping M4 as the final determination result for whether a cooling water leak has occurred. Then, the control device 290 causes the process to proceed to step S14.

[0045] In step S14, the control device 290 outputs a signal indicating the final determination result of the determination process to the display 220. As a result, the final determination result of the determination process is displayed on the display 220. The display 220 is an example of prescribed hardware used to notify the operator of the final determination result.

[0046] <First Temporary Judgment Control>

[0047] Next, the first temporary determination control in step S12 of the anomaly determination device 200 will be described. In this embodiment, when the control device 290 of the anomaly determination device 200 starts the process of step S12, it first executes the first temporary determination control.

[0048] like Figure 4 As shown, when the control device 290 of the anomaly determination device 200 starts the first temporary determination control, it performs the processing in step S21. In step S21, the control device 290 reads the image data captured in step S11. Afterwards, the control device 290 causes the processing to proceed to step S22.

[0049] In step S22, the control device 290 converts the image data read in step S21 to grayscale. Specifically, the control device 290 converts the image of each pixel in the image data read in step S21 into an image of gray pixels that vary between white and black. Then, the control device 290 causes the processing to proceed to step S23.

[0050] In step S23, the control device 290 performs high-pass filtering on the image data processed in step S22. Specifically, the control device 290 attenuates the low-frequency components contained in the image data of step S22, making the white parts more prominent. Then, the control device 290 proceeds to step S24.

[0051] In step S24, the control device 290 performs binarization processing on the image data processed in step S23. Specifically, the control device 290 converts the image of each pixel in step S23 into an image of white or black pixels. Additionally, the control device 290 performs dilation filtering on the image data. Specifically, the control device 290 expands the white portion by converting the black pixels surrounding the white pixels to white. Afterwards, the control device 290 proceeds to step S25.

[0052] In step S25, the control device 290 extracts white pixels from the image data processed in step S24. Then, the control device 290 obtains the number of white pixels. Here, the portion of the outer wall surface of the water pump 20 where liquid cooling water adheres is prone to appearing white due to light reflection. Therefore, the processing in step S25 is to obtain the number of pixels based on the area of ​​the region in the image data where liquid cooling water is highly likely to adhere. Afterwards, the control device 290 proceeds to step S26.

[0053] In step S26, the control device 290 extracts from the image data read in step S21 the portion of the outer wall surface of the water pump 20 where cooling water adheres and the portion containing components of the cooling water that precipitates out. As described above, since cooling water is pink, the portion of the image data in step S21 where cooling water adheres is pink. Furthermore, if there is a leakage of cooling water within a design-permissible range, the cooling water rapidly vaporizes. At this time, the components contained in the cooling water adhere to the outer wall surface of the water pump 20 as pink precipitates. Therefore, the processing in step S26 is to extract the pink pixels in the image data of step S21 as the portion where cooling water and precipitates adhere. Afterwards, the control device 290 causes the processing to proceed to step S27.

[0054] In step S27, the control device 290 obtains the total number of pink pixels extracted in step S26. Therefore, the processing in step S27 is a process of obtaining the number of pixels by taking the area of ​​the portion of the image data read in step S21 with which cooling water and precipitates are attached. In this embodiment, the processing in steps S25 and S27 is an acquisition process. Afterwards, the control device 290 causes the processing to proceed to step S28.

[0055] In step S28, the control device 290 inputs the number of white pixels in step S25 and the total number of pink pixels in step S27 as input variables to the first mapping M1 defined by the mapping data 294A. The first mapping M1 outputs the white point ratio WR as an output variable indicating whether a cooling water leak has occurred in the water pump 20. Specifically, the first mapping M1 outputs the white point ratio WR according to the following formula (1).

[0056] Formula (1): White point ratio WR = (number of white pixels in step S25) / (total number of pink pixels in step S27) × 100.

[0057] In this embodiment, step S28 is a calculation process. Afterwards, the control device 290 causes the process to proceed to step S29.

[0058] In step S29, the control device 290 determines whether the white spot ratio WR is greater than or equal to a predetermined first threshold Z1. Furthermore, a large white spot ratio WR indicates that, relative to the overall portion of the outer wall surface of the water pump 20 where cooling water and precipitates are attached, the proportion of liquid cooling water is large. Here, the first threshold Z1 is determined, for example, as follows: First, the sealing component 25 is deteriorated by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20 is acquired. Furthermore, based on the acquired image data, the white spot ratio WR is calculated according to the procedures of steps S21 to S28. Moreover, during the deterioration process, a skilled operator determines whether a cooling water leak has occurred in the water pump 20. Then, based on the white spot ratio WR and the skilled operator's determination, the first threshold Z1 is set. If the white spot ratio WR is greater than or equal to the first threshold Z1, the control device 290 temporarily determines that a cooling water leak has occurred in the water pump 20. On the other hand, if the white spot ratio WR is less than the first threshold Z1, the control device 290 temporarily determines that no cooling water leakage has occurred in the water pump 20. Therefore, the processing in step S29 is the first temporary determination processing. Afterwards, the control device 290 ends the first temporary determination control.

[0059] <Second Temporary Judgment Control>

[0060] Next, the second temporary determination control using the k-nearest neighbor method in step S12 of the anomaly determination device 200 will be described. In this embodiment, the anomaly determination device 200 executes the second temporary determination control after the first temporary determination control.

[0061] like Figure 5As shown, when the control device 290 of the anomaly determination device 200 starts the second temporary determination control, it performs the processing in step S41. In step S41, the control device 290 reads the image data captured in step S11. Afterwards, the control device 290 causes the processing to proceed to step S42.

[0062] In step S42, the control device 290 extracts the portion of the image data read in step S41 that contains attached cooling water and precipitates. This step S42 is the same as the process described in step S26. Afterwards, the control device 290 proceeds to step S43.

[0063] In step S43, the control device 290 segments the portion containing the cooling water and precipitates extracted in step S42 into multiple images. At this time, the control device 290 performs the segmentation process in a manner that ensures the number of pixels in each segmented image is equal to the number of pixels in the segmented images. Then, the control device 290 proceeds to step S46.

[0064] In step S46, the control device 290 acquires the hue H for each segmented image from step S43. In other words, the process in step S46 is to acquire the hue H for each part with adhering cooling water and precipitates. Afterwards, the control device 290 causes the process to proceed to step S47.

[0065] In step S47, the control device 290 obtains the saturation S for each segmented image from step S43. In other words, the process in step S47 is to obtain the saturation S for each part with adhering cooling water and precipitates. In this embodiment, the processes in steps S46 and S47 are acquisition processes. Afterward, the control device 290 causes the process to proceed to step S48.

[0066] In step S48, the control device 290, for each segmented image from step S43, inputs the hue H from step S46 and the saturation S from step S47 as input variables to the second mapping M2 defined by the mapping data 294A. The second mapping M2 outputs an output variable indicating whether a cooling water leak has occurred in the water pump 20. Therefore, in step S48, the control device 290, for each segmented image from step S43, obtains an output variable indicating whether a cooling water leak has occurred in the water pump 20. Furthermore, if a cooling water leak has occurred in the water pump 20, the output variable of the second mapping M2 is "1". Conversely, if no cooling water leak has occurred in the water pump 20, the output variable of the second mapping M2 is "0". In this embodiment, the processing in step S48 is a calculation process.

[0067] Here, the second mapping M2 is determined, for example, as follows. First, the sealing component 25 is degraded by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20 is acquired. Furthermore, based on the acquired image data, the hue H and saturation S are acquired according to the steps S41 to S47. Then, as an example, the k-nearest neighbor method is used to classify the data into two groups based on hue H and saturation S, thereby enabling the second mapping M2 to learn. Furthermore, based on the deterioration condition of the sealing component 25, one of the two groups is designated as the group where cooling water leakage has occurred in the water pump 20. Further, the other of the two groups is designated as the group where cooling water leakage has not occurred in the water pump 20. After step S48, the control device 290 causes the processing to proceed to step S51.

[0068] In step S51, the control device 290 obtains the number of images from the segmented images obtained in step S43 showing a cooling water leak in the water pump 20. Then, the control device 290 causes the processing to proceed to step S52.

[0069] In step S52, the control device 290 obtains the total number of segmented images from step S43. Therefore, the processing in step S52 is a process of obtaining the total number of segmented images as a portion with attached cooling water and precipitates. Afterwards, the control device 290 causes the processing to proceed to step S53.

[0070] In step S53, the control device 290 calculates the leakage ratio LR based on the number of images showing cooling water leakage in step S51 and the total number of segmented images in step S52. Furthermore, the leakage ratio LR is expressed by the following equation (2).

[0071] Equation (2): Leakage ratio LR = (number of images in step S51 in which cooling water leakage occurred) / (total number of segmented images in step S52) × 100.

[0072] Then, the control device 290 causes the process to proceed to step S54.

[0073] In step S54, the control device 290 determines whether the leakage ratio LR is greater than or equal to a predetermined second threshold Z2. Here, the second threshold Z2 is determined, for example, as follows: First, the sealing component 25 is degraded by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20 is acquired. Furthermore, based on the acquired image data, the leakage ratio LR is calculated according to the procedures of steps S41 to S53. Moreover, during the deterioration process, a skilled operator determines whether a cooling water leak has occurred in the water pump 20. Then, based on the leakage ratio LR and the skilled operator's determination, the second threshold Z2 is set. If the leakage ratio LR is greater than or equal to the second threshold Z2, the control device 290 temporarily determines that a cooling water leak has occurred in the water pump 20. On the other hand, if the leakage ratio LR is less than the second threshold Z2, the control device 290 temporarily determines that no cooling water leak has occurred in the water pump 20. Therefore, the process in step S54 is a second temporary determination process. Afterwards, control device 290 terminates the second temporary determination control.

[0074] <Third Temporary Judgment Control>

[0075] Next, the third temporary determination control using a support vector machine in step S12 of the anomaly determination device 200 will be described. In this embodiment, the anomaly determination device 200 performs the third temporary determination control after the second temporary determination control. Furthermore, a portion of the processing of the third temporary determination control is the same as a portion of the processing of the second temporary determination control. Therefore, in the description of the third temporary determination control, the same reference numerals are used for the processes common to the second temporary determination control, and the description is omitted or simplified.

[0076] like Figure 6 As shown, when the control device 290 of the anomaly determination device 200 starts the third temporary determination control, it performs the processing of step S41. Then, the control device 290 executes the processing of steps S41 to S47. After step S47, the control device 290 causes the processing to proceed to step S68.

[0077] In step S68, the control device 290 inputs the hue H from step S46 and the saturation S from step S47 as input variables to the third mapping M3 defined by the mapping data 294A for each segmented image from step S43. The third mapping M3 outputs an output variable indicating whether a cooling water leak has occurred in the water pump 20. Therefore, in step S68, the control device 290 obtains an output variable indicating whether a cooling water leak has occurred in the water pump 20 for each segmented image from step S43. Furthermore, if a cooling water leak has occurred in the water pump 20, the output variable of the third mapping M3 is "1". Conversely, if no cooling water leak has occurred in the water pump 20, the output variable of the third mapping M3 is "0". In this embodiment, the processing in step S68 is a calculation process.

[0078] Here, the third mapping M3 is determined, for example, as follows. First, the sealing component 25 is degraded by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20 is acquired. Furthermore, based on the image data, the hue H and saturation S are obtained according to steps S41 to S47. Then, as an example, the third mapping M3 is learned using a support vector machine to classify the data into two groups based on hue H and saturation S. Furthermore, based on the deterioration condition of the sealing component 25, one of the two groups is set as the group where cooling water leakage has occurred in the water pump 20. Further, the other of the two groups is set as the group where cooling water leakage has not occurred in the water pump 20.

[0079] After step S68, control device 290 causes the process to proceed to step S51. Then, control device 290 executes the processes from steps S51 to S53. After step S53, control device 290 causes the process to proceed to step S74.

[0080] In step S74, the control device 290 determines whether the leakage ratio LR is greater than or equal to a predetermined third threshold Z3. Here, the third threshold Z3 is determined, for example, as follows: First, the sealing component 25 is degraded by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall of the water pump 20 is acquired. Furthermore, based on the image data, the leakage ratio LR is calculated according to the procedures of steps S41 to S53. Moreover, during the deterioration process, a skilled operator determines whether a cooling water leak has occurred in the water pump 20. Then, based on the leakage ratio LR and the skilled operator's determination, the third threshold Z3 is set. If the leakage ratio LR is greater than or equal to the third threshold Z3, the control device 290 temporarily determines that a cooling water leak has occurred in the water pump 20. On the other hand, if the leakage ratio LR is less than the third threshold Z3, the control device 290 temporarily determines that no cooling water leak has occurred in the water pump 20. Therefore, the process in step S74 is a third temporary determination process. Afterwards, control device 290 terminates the third temporary determination control.

[0081] <Fourth Temporary Judgment Control>

[0082] Next, the fourth temporary determination control in step S12 of the anomaly determination device 200 will be described. In this embodiment, the anomaly determination device 200 executes the fourth temporary determination control after the third temporary determination control.

[0083] like Figure 7 As shown, when the control device 290 of the anomaly determination device 200 starts the fourth temporary determination control, it performs the processing in step S81. In step S81, the control device 290 reads the image data captured in step S11. Afterwards, the control device 290 causes the processing to proceed to step S82.

[0084] In step S82, the control device 290 extracts the portion of the image data read in step S81 that contains attached cooling water and precipitates. This step S82 is the same as the process described in step S26. Afterwards, the control device 290 proceeds to step S83.

[0085] In step S83, the control device 290 obtains a hue H for each pixel containing the portion with attached cooling water and precipitates extracted in step S82. Then, the control device 290 obtains an average hue HA as the average of all obtained hues H. Afterward, the control device 290 causes the process to proceed to step S84.

[0086] In step S84, the control device 290 obtains a saturation S for each pixel containing the portion with attached cooling water and precipitates extracted in step S82. Then, the control device 290 obtains an average saturation SA as the average of all obtained saturations S. Afterward, the control device 290 causes the process to proceed to step S85.

[0087] In step S85, the control device 290 obtains the luminance (brightness) V for each pixel containing the portion with attached cooling water and precipitates extracted in step S82. Then, the control device 290 obtains the average luminance VA as the average of all obtained luminance V values. Afterward, the control device 290 causes the process to proceed to step S86.

[0088] In step S86, the control device 290 obtains the number of pixels containing the portions of cooling water and precipitates extracted in step S82. Additionally, the control device 290 obtains the total number of pixels in the image data read in step S81. Then, based on the number of pixels containing the portions of cooling water and precipitates extracted in step S82 and the total number of pixels in the image data read in step S81, the control device 290 calculates the precipitation ratio CR. Furthermore, the precipitation ratio CR is expressed by the following equation (3).

[0089] Equation (3): Extraction ratio CR = (number of pixels with cooling water and precipitates extracted in step S82) / (total number of pixels of image data read in step S81) × 100.

[0090] Then, the control device 290 causes the process to proceed to step S87.

[0091] In step S87, the control device 290 obtains the white point ratio WR calculated in step S28 of the first temporary determination control. Then, the control device 290 causes the processing to proceed to step S88.

[0092] In step S88, the control device 290 extracts white pixels from the image data processed in step S24, which underwent the first temporary determination control. Then, the control device 290 obtains the size of the block of white pixels as the white point area. For example, if a total of 10 white pixels are arranged in a block that is vertically and horizontally adjacent, the control device 290 sets the white point area to "10". Alternatively, for example, if multiple white pixels are not adjacent and there is only one white pixel in each block, the control device 290 sets the white point area to "1". The control device 290 determines the white point area for all white pixels. Then, the control device 290 obtains the average white point area WA as the average of all white point areas. In this embodiment, the processing in steps S83 to S88 is an acquisition process. Afterwards, the control device 290 causes the processing to proceed to step S91.

[0093] In step S91, the control device 290 inputs the average hue HA, average saturation SA, average brightness VA, precipitation ratio CR, white point ratio WR, and average white point area WA as input variables, and inputs the fourth mapping M4 defined by the mapping data 294A. The fourth mapping M4 outputs an output variable indicating whether a cooling water leak has occurred in the water pump 20. Furthermore, if a cooling water leak has occurred in the water pump 20, the output variable of the fourth mapping M4 is "1". Conversely, if no cooling water leak has occurred in the water pump 20, the output variable of the fourth mapping M4 is "0". In this embodiment, the processing in step S91 is a calculation process. Furthermore, the fourth mapping M4 is the mapping with the most types of variables used as input variables among the first mapping M1 to the fourth mapping M4. Therefore, in this embodiment, the fourth mapping M4 is a specific mapping.

[0094] Here, the fourth mapping M4 is determined, for example, as follows. First, the sealing component 25 is degraded by driving the internal combustion engine 100 through experiments, etc. During this deterioration process, image data of the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20 is acquired. Furthermore, based on the image data, the average hue HA, average saturation SA, average brightness VA, precipitation ratio CR, white point ratio WR, and average white point area WA are acquired according to the steps S81 to S88. Then, as an example, the k-nearest neighbor method is used to classify the data into two groups based on the average hue HA, average saturation SA, average brightness VA, precipitation ratio CR, white point ratio WR, and average white point area WA, and the fourth mapping M4 is learned. Furthermore, based on the deterioration condition of the sealing component 25, one of the two groups is set as the group in which cooling water leakage has occurred in the water pump 20. Further, the other of the two groups is set as the group in which cooling water leakage has not occurred in the water pump 20. In this embodiment, the first mapping M1 to the fourth mapping M4 are mappings pre-learned through ensemble learning based on the same learning data obtained under the same conditions during the process of deteriorating the sealing component 25 as described above. After step S91, the control device 290 causes the process to proceed to step S92. Furthermore, ensemble learning can be described as a method to improve the predictive accuracy of machine learning, for example, by combining multiple models. Examples of ensemble learning include bagging, boosting, and stacking.

[0095] In step S92, the control device 290 temporarily determines whether a cooling water leak has occurred in the water pump 20 based on the output variable of step S91. Therefore, the processing in step S92 is the fourth temporary determination process. Afterwards, the control device 290 ends the current fourth temporary determination control.

[0096] <The function of this implementation method>

[0097] In the water pump 20 of the internal combustion engine 100, due to deterioration of the sealing component 25, liquid cooling water sometimes leaks through the sealing component 25 from the flow space 100Z to the through hole 21A. As a result, the liquid cooling water leaks to the outside of the pump housing 21 through the lower passage 21E and the lower space 21D. Thus, in the case of liquid cooling water leakage, the amount of cooling water leaked is relatively large. Furthermore, in the case of liquid cooling water leakage, a relatively large amount of cooling water adheres to the periphery of the opening of the lower space 21D in the outer wall surface of the water pump 20. That is, the periphery of the opening of the lower space 21D becomes wet.

[0098] When determining whether a cooling water leak has occurred in the water pump 20, the operator uses the camera 210 of the anomaly detection device 200 to capture images of the periphery of the opening in the lower space 21D on the outer wall of the water pump 20 (S11). Furthermore, the control device 290 of the anomaly detection device 200 performs a total of four temporary determination processes using the first mapping M1 to the fourth mapping M4 based on the captured image data (S12). Then, based on the temporary determination results of the four temporary determination processes, the control device 290 performs a final determination process to determine whether a cooling water leak has occurred (S13).

[0099] <Effects of this implementation method>

[0100] (1) In this embodiment, there is no room for subjective intervention by the operator in the series of determinations. Therefore, for example, the determination result will not be biased by the subjectivity of each operator performing the maintenance of the water pump 20.

[0101] (2) In the determination process, the control device 290 uses the temporary determination results from the four temporary determination processes (first mapping M1 to fourth mapping M4) that account for more than half of the temporary determination results as the final determination result for whether a cooling water leak has occurred. Therefore, for example, compared to the case where the final determination result is obtained based solely on the temporary determination result from any one of the first to fourth temporary determination processes, the accuracy of the determination result can be ensured.

[0102] (3) In this embodiment, the fourth mapping M4 is the mapping with the most types of variables used as input variables among the first mappings M1 to the fourth mapping M4. Therefore, the more input variables are input to the fourth mapping M4, the higher the reliability of the temporary determination result of the fourth temporary determination process based on the output variables output from the fourth mapping M4. Considering this, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that no cooling water leak has occurred, the control device 290 uses the temporary determination result of the fourth temporary determination process using the fourth mapping M4 as the final determination result of whether a cooling water leak has occurred. Therefore, even when the number of different temporary determination results is conflicting, the result of the fourth temporary determination process, which is considered more accurate, can be used as the final determination result.

[0103] (4) In this embodiment, the first mapping M1 to the fourth mapping M4 are mappings that have been pre-learned through integrated learning based on the same learning data obtained under the same conditions during the process of deteriorating the sealing component 25. Therefore, for example, compared with the case where multiple mappings are learned based on learning data obtained under different conditions, it is possible to suppress the deviation of the determination result.

[0104] <Variation Example>

[0105] This embodiment can be implemented with the following modifications. This embodiment and the following variations can be combined and implemented with each other within the scope of technical inconsistency.

[0106] • In the above embodiments, the input variables for the mapping specified by the mapping data 294A can be changed.

[0107] For example, as input variables for the second mapping M2, hue H and saturation S can be replaced, or luminance V can be used in addition to hue H and saturation S. Similarly, the input variables for the third mapping M3 can be changed. Furthermore, as input variables for the fourth mapping M4, some input variables can be omitted, or other input variables can be added.

[0108] For example, the input variables for the second mapping M2 may not be the same as those for the third mapping M3. Additionally, for example, the number of variables used as input variables for the second mapping M2 may be greater or less than the number of variables used as input variables for the third mapping M3.

[0109] For example, the number of types of variables used as input variables in the second mapping M2 can be greater than the number of types of variables used as input variables in the fourth mapping M4. In this case, if the second mapping M2 has the most types of variables used as input variables among the first mappings M1 to the fourth mappings M4, then the second mapping M2 is a specific mapping. That is, the specific mapping that takes precedence when the same number of temporary decision results conflict with each other in the decision-making process may not be the fourth mapping M4.

[0110] In the above embodiment, the number of temporary decision-making processes performed by the control device 290 can be two or more, three or fewer, or five or more. In this case, the mapping data 294A only needs to specify the mapping corresponding to the number of temporary decision-making processes.

[0111] In the above embodiment, when the number of provisional determinations indicating a cooling water leak is the same as the number of provisional determinations indicating no cooling water leak, the control device 290 may retain the final determination. Furthermore, in this case, for example, it is preferable that the control device 290 outputs a signal to the display 220 to cause the display 220 to show a message urging the operator to re-determine the issue.

[0112] In the above embodiments, the first mapping M1 to the fourth mapping M4 may not be based on the same learning data, but rather on mappings that have been pre-learned through ensemble learning. For example, the first mapping M1 to the fourth mapping M4 may be learned based on learning data obtained under different conditions. If the first mapping M1 to the fourth mapping M4 can be learned based on a large amount of learning data, the possibility of bias in the judgment results is low.

[0113] • In the above embodiments, the configuration of the first mapping M1 to the fourth mapping M4 can also be changed.

[0114] For example, examples are given of second mapping M2 and third mapping M3 using the k-nearest neighbor method or support vector machines, but these are not limited to. As a specific example, second mapping M2 and third mapping M3 can also use neural networks.

[0115] In the above embodiments, the execution device is not limited to an execution device having a control circuit with a CPU 291 and a ROM 293 that performs software processing. As a specific example, it may also include a dedicated hardware circuit, such as an ASIC, that performs hardware processing on at least a portion of the software processing performed in the above embodiments. That is, the execution device can be configured as any one of (a) to (c) below: (a) Includes: a processing device that executes all of the above processing according to a program; and a program storage device (including a non-transitory computer-readable storage medium) such as a ROM that stores the program. (b) Includes: a processing device that executes a portion of the above processing according to a program and a program storage device; and a dedicated hardware circuit that executes the remaining processing. (c) Includes a dedicated hardware circuit that executes all of the above processing. Here, there may be multiple software execution devices and dedicated hardware circuits, including processing devices and program storage devices.

[0116] • In the above embodiments, the structure of the anomaly determination device 200 can be changed.

[0117] For example, an anomaly detection device 200 can be a server. Specifically, the process can be performed as follows: In step S11 described above, the smartphone sends image data acquired by the smartphone to the server via a communication network. The server then performs steps S12 and S13. Further, the server sends a signal indicating the final determination of step S13 to the smartphone via the communication network. The smartphone then performs step S14. In this modified example, the anomaly detection device 200 installed on the server does not include a camera 210 or a display 220. That is, the anomaly detection device 200 only needs an execution device and a storage device. Devices for capturing image data and outputting determination results are not necessary for the anomaly detection device 200.

[0118] • In the above embodiments, the structure of the internal combustion engine 100 can be changed.

[0119] For example, water pump 20 is exemplified as a so-called mechanical pump driven by a driving force from the crankshaft of internal combustion engine 100, but it is not limited to this. Specifically, water pump 20 may also be a so-called electric pump driven by a driving force from an electric motor.

Claims

1. An anomaly detection device, specifically an anomaly detection device for a water pump, wherein, The water pump that supplies cooling water to the internal combustion engine is the object of the anomaly detection device. The anomaly detection device includes an execution circuit and a storage circuit. The storage circuit is configured to store mapping data of a specified mapping, and is configured to input variables into the mapping, and output an output variable indicating whether a leakage of cooling water has occurred in the water pump. The execution circuit is configured to perform the following processing: The acquisition process involves obtaining the input variables from image data obtained by capturing images of the outer surface of the water pump. The calculation process outputs the value of the output variable by inputting the input variable obtained by the acquisition process into the mapping. The temporary determination process involves performing a temporary determination, based on the output variable, to determine whether a leak of cooling water has occurred in the water pump. and The determination process involves determining, based on the provisional determination result (which serves as the provisional determination result), whether a leak of cooling water has occurred in the water pump. The mapping is one of a plurality of mutually distinct mappings defined by the mapping data. One or more of the aforementioned mappings were learned in advance via machine learning. The temporary decision-making process is one of several temporary decision-making processes. The provisional determination result is one of multiple provisional determination results. The execution circuit is configured to perform the computational process for each of the plurality of mappings and obtain a plurality of temporary determination results, each of the plurality of temporary determination results being obtained by performing the temporary determination process for each of the output variables output from the plurality of mappings. The execution circuit is configured such that, in the determination process, more than half of the temporary determination results are taken as the final determination result of whether the cooling water leakage has occurred.

2. The anomaly determination device according to claim 1, wherein, The mapping that uses the most types of variables as input variables among the multiple mappings is a specific mapping. The execution circuit is configured such that, in the determination process, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that a cooling water leak has not occurred, the temporary determination result when the temporary determination process was performed based on the output variable output from the specific mapping is taken as the final determination result of whether a cooling water leak has occurred.

3. The anomaly determination device according to claim 1, wherein, The multiple mappings were pre-learned through ensemble learning based on the same learning data.

4. An anomaly detection method, specifically a method for detecting anomalies in a water pump, wherein, The storage circuit stores mapping data with a specified mapping. The anomaly detection method includes the following steps: The execution circuit obtains the input variable for the mapping from the image data obtained by photographing the outer surface of the water pump, and the water pump pressurizes the cooling water of the internal combustion engine; The execution circuit performs a calculation process that outputs the value of the output variable of the mapping by inputting the input variable into the mapping. The output variable indicates whether a leak of cooling water has occurred in the water pump. The execution circuit performs a temporary determination process, which is used to obtain a temporary determination result based on the output variable to determine whether a leak of cooling water has occurred in the water pump; and The execution circuit performs a determination process, which, based on the provisional determination result, yields a final determination result as to whether a leak of cooling water has occurred in the water pump. The mapping is one of a plurality of mutually distinct mappings defined by the mapping data. One or more of the aforementioned mappings were learned in advance via machine learning. The temporary decision-making process is one of several temporary decision-making processes. The provisional determination result is one of multiple provisional determination results. The anomaly detection method further includes: The execution circuit performs the computational processing for each of the plurality of mappings; The execution circuit obtains a plurality of temporary determination results, each of which is obtained by performing the temporary determination process on each of the output variables output from the plurality of mappings; and In the determination process, the execution circuit takes more than half of the temporary determination results as the final determination result.

5. The anomaly determination method according to claim 4, wherein, The mapping that uses the most types of variables as input variables among the multiple mappings is a specific mapping. The anomaly detection method further includes the following steps: In the determination process, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that no cooling water leak has occurred, the execution circuit uses the temporary determination result obtained when the temporary determination process was performed based on the output variable output from the specific mapping as the final determination result of whether a cooling water leak has occurred.

6. A non-transitory computer-readable storage medium storing a program that causes an execution circuit to perform an abnormality determination process for a water pump, wherein, The storage circuit stores mapping data with a specified mapping. The anomaly detection and processing includes the following steps: The execution circuit obtains the input variable for the mapping from the image data obtained by capturing the outer surface of the water pump, and the water pump pressurizes the cooling water of the internal combustion engine; The execution circuit performs a calculation process that outputs the value of the output variable of the mapping by inputting the input variable into the mapping. The output variable indicates whether a leak of cooling water has occurred in the water pump. The temporary determination process is executed by the execution circuit. This temporary determination process is used to obtain a temporary determination result based on the output variable to determine whether a leak of cooling water has occurred in the water pump. and The execution circuit performs a determination process, which, based on the provisional determination result, yields a final determination result as to whether a leak of cooling water has occurred in the water pump. The mapping is one of a plurality of mutually distinct mappings defined by the mapping data. One or more of the aforementioned mappings were learned in advance via machine learning. The temporary decision-making process is one of several temporary decision-making processes. The provisional determination result is one of multiple provisional determination results. The anomaly detection and processing also includes the following steps: The execution circuit performs the computational processing for each of the plurality of mappings; The execution circuit obtains a plurality of temporary determination results, each of which is obtained by performing the temporary determination process on each of the output variables output from the plurality of mappings; and In the determination process, the execution circuit takes more than half of the temporary determination results as the final determination result.

7. The non-transitory computer-readable storage medium according to claim 6, wherein, The mapping that uses the most types of variables as input variables among the multiple mappings is a specific mapping. The anomaly detection and processing also includes the following steps: In the determination process, when the number of temporary determination results indicating that a cooling water leak has occurred is the same as the number of temporary determination results indicating that no cooling water leak has occurred, the execution circuit uses the temporary determination result obtained when the temporary determination process was performed based on the output variable output from the specific mapping as the final determination result of whether a cooling water leak has occurred.

Images